Download Free CE

Sovereign, air-gapped agentic AI

A fully self-hosted, air-gappable AI-native platform for government and defense: in-database agents, grounded RAG, property-graph analysis, in-database ML, and a tamper-evident immutable audit — all offline, no external AI dependency, no data egress.

Request a briefingSee it runSelf-host the free CE

Evaluating for a disconnected or classified environment? Self-host the CE to prototype, then request a sovereign-deployment briefing.

The pain we built this for

If you deliver capability in the public sector, the AI story stalls at the boundary: “The mission needs modern AI, but the data can’t leave the enclave, there can’t be an external model dependency, and every automated action has to be accountable.”

  • Controlled and classified data cannot egress to a cloud model — which removes almost every commercial AI offering.
  • An external model dependency is a supply-chain and availability risk; the capability has to run disconnected.
  • Standing up separate systems for search, graph, ML, and audit multiplies the accreditation and attack surface.
  • Every automated action needs an accountable, reconstructable record for oversight and after-action review.
  • Autonomy has to be bounded — an agent that can act without a human check is not acceptable.

See it — the whole stack, inside the boundary

A bounded agent and a tamper-evident audit — running with no external AI dependency, in a few statements.

-- One engine, inside the boundary: agents + RAG + graph + immutable audit.
CREATE AGENT analyst
  PERSONA 'Answer only from cited records inside this deployment. Never
           assert beyond the evidence. Surface links for a human analyst.'
  ALLOW_WRITES FALSE;

-- Accountable, tamper-evident record of every action — inside the enclave.
CREATE IMMUTABLE TABLE agent_audit (
  task_id   TEXT,
  actor     TEXT,
  sources   JSONB,
  output    TEXT,
  reviewer  TEXT
) WITH (ENCRYPTION = 'AES256GCM');

VERIFY TABLE agent_audit;             -- prove the trail was never altered

Six capabilities. One air-gappable engine.

Agents, RAG, graph, ML, and immutable audit in a single deployable — one accreditation, one boundary, offline.

Fully air-gapped inference

Local models run inside the enclave — embeddings, generation, agents, and ML with no external call. Mirror models and recipes in once; run with zero internet dependency.

Local inference · offline mirror
In-database agents

Agentic analysis over your own data, bounded by an allowlist and allow_writes controls — automation that reasons, without an external AI dependency or an autonomous write it should not have.

CREATE AGENT · guardrails
Grounded RAG on your corpus

Retrieval-grounded answers with citations over classified or controlled documents — grounded to source, inside the boundary.

Vector RAG · citations · EMBED
Graph & entity analysis

Model entities, relationships, and events as a property graph and traverse them in Cypher — link analysis and impact assessment over one store.

Property graph · Cypher · GraphRAG
Tamper-evident immutable audit

Every action writes to a hash-chained immutable table; VERIFY TABLE proves integrity — an accountable, reconstructable trail for oversight inside the enclave.

CREATE IMMUTABLE TABLE · VERIFY TABLE
In-database ML & anomaly detection

Train and score models where the data lives — anomaly detection and prediction without moving controlled data to a separate ML platform.

AutoML · anomaly detection · SQL

The objections an ISSO raises first

Data egress & sovereignty

Nothing leaves the deployment — local inference, one self-hosted engine, no cloud model or hosted vector store in the path. You own the boundary and the keys.

Air-gap & offline

Mirror models and recipes in once; run with zero internet dependency. Deployable in disconnected and classified environments.

Accountability

A tamper-evident, reconstructable audit of every agent action and decision, verifiable with VERIFY TABLE, for oversight and after-action review.

One engine, fewer components

SQL, vectors, graph, ML, agents, and audit in a single deployable — a smaller footprint and a smaller attack surface than a stitched stack.

Run it yourself — the sovereign-stack recipe library

Every capability above is a runnable recipe — immutable audit, self-hosted agent backend, graph correlation. Certified against the latest release.

Tamper-evident agent audit
Accountable AI inside the enclave, verified
Entity link analysis
Who connects to whom, in one graph query
Grounded offline retrieval
Local inference, no egress, cited sources
Full self-hosted agent backend
Memory + RAG + graph + audit in SQL
Entity / threat correlation graph
Link analysis over one store
Browse the full recipe library →
Every recipe is runnable, certified on the latest release

Why one engine beats a stitched stack

  • Agents, RAG, graph, ML, and audit in one deployable — one accreditation boundary and a smaller attack surface than four integrated systems.
  • Local inference means no external model dependency and no egress — deployable disconnected.
  • Recipes are SQL and Cypher you can read, review, and adapt — accountable automation, not an opaque service.

5-email course · free

Ship an agent with persistent memory in 5 emails

A 5-email course where you build a real agent with persistent memory — one runnable SQL recipe per email, no LangChain, no Pinecone, no Redis. Lesson 1 hits your inbox the second you sign up. After that: a new recipe every Monday. No spam.

Free forever. We'll never share your email.

Modern AI that never leaves the enclave

Agents, RAG, graph, ML, and a tamper-evident audit — one air-gappable engine, no external dependency, no egress. Request a sovereign-deployment briefing, or self-host the free CE.

Request a briefingSelf-host the free CEStar on GitHub